Résumé
PREPRINT // SUBMITTED THE 15TH OF OCTOBER 2020 TO ANNALS OF GIS // SPECIAL ISSUE "GENERAL PRINCIPLES/ANALYTICAL FRAMEWORKS IN GEOGRAPHY/GISCIENCE" The current biodiversity crisis, combined with climate change are major issues requesting specific monitoring of plants communities' responses in terms of geographical distribution. Nowadays, large open-access biodiversity datasets (OBDs) such as the Global Biodiversity Information Facility (GBIF) are commonly used to describe, explain and predict fauna and flora geographical distribution. They constitute new opportunities, but stay related to major uncertainties about sampling biases, driven by the concentration of various biodiversity data records and associated data providers. Taking the example of a widely studied tree (Betula pubsescens Ehrh.), in a scientifically well-funded region (Swedish Lapland), we discuss those peculiar issues in the frame of geographical laws (e.g. respectively spatial autocorrelation, heterogeneity and similarity) at macro-and micro-regional scales. After spatial and temporal filtering on georeferenced records and discussion on sampling strategies heterogeneity, tests of spatial autocorrelation (Moran's I index) has been conducted on Birch tree records provided by major institutions, comparatively. Pearson Khi-2 (χ 2) test is thus applied on the generated grid to confront number of Birch tree records with accessibility factors (e.g. artificial land cover, roads, protected natural areas) Thus, a micro-regional analysis is conducted to quantify Birch tree records in vegetation classes where this tree species is supposed to be dominant. At the macro-regional scale, results show the high spatial variability of sub-datasets according to institution providing records from the studied GBIF OBDs (with higher autocorrelation results for large contributors). This spatial variability, and high spatial autocorrelation effects appears to be partly explained at macro-, micro-and local scale by the distribution of human accessibility and facility factors (e.g. roads, cities etc). In this study case, exploring OBDs for an extensively addressed tree species, in a significantly funded region/country was particularly useful, demonstrating the relevance of spatial similarity law to differentiate adequately sampling biases efforts from "natural" spatial autocorrelation.